AI is not yet driving drug development
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Aïda Amer/Axios. Stock: Getty Images
Curing disease is one of AI's most tantalizing use cases, fetching billions of VC dollars, but thus far it's generated more excitement than evidence.
- That's the thrust of a peer-reviewed paper published in Nature Reviews Drug Discovery, which calls the clinical impact of AI "disappointingly limited."
The big picture: The researchers find that while AI has gotten good at identifying drug candidates, it hasn't done much to prove those predictions will survive the complexity of human biology.
- Instead, AI-derived drugs still run into the same Phase II bottleneck as conventional drugs: Proving efficacy in large, diverse patient populations, where failure rates are high and costs are astronomical.
Zoom in: This mostly boils down to cellular data sets, an unsexy thing to devote resources to when you could instead be swinging for blockbuster treatments. In some cases, the data isn't robust enough. In most cases, it's just too messy.
- "We don't know how to clean this stuff up and categorize it for real ML/AI, and honestly, we don't even know if it can be," writes biotech journalist Derek Lowe when discussing the research paper.
State of play: Venture capitalists have invested more than $40 billion into AI biopharma this decade, per PitchBook, but there are precious few medicines that have moved beyond Phase II.
- Yes, drug development is a lengthy process. But, as the researchers note, "many AI proponents have claimed these timelines will be meaningfully shortened."
Yes, but: This isn't to say there are no current late-stage applications.
- The most notable may be a new Phase III trial by Moderna and Merck, which is using AI to choose tumor targets in cancer patients — and then treating them with an existing medicine.
- But, again, this is more about decision-making for specific individuals than it is wholesale drug development.
What they're saying: Axios reached out to some top biotech VCs to get their thoughts. Two notable replies:
- Robert Nelsen, Arch Venture Partners: "We need more data for models of biology. We need regulatory reform to move faster. And when we have the cellular data and more population data, and eventually quantum computing, it will fundamentally disrupt the industry."
- Bijan Salehizadeh, NaviMed Capital: "The biggest costs are in the D of R&D, where Phase 2 and Phase 3 costs are massive and inflated by CROs and baked-in inefficiencies and regulation ... I'm afraid the AI applications in development will be surface level, focused on low-hanging fruit like biostats and data lock once a trial is enrolled and medical writing for FDA submissions. But not the big stuff of how to enroll a phase 2 or phase 3 program at hundreds to thousands of global sites and do all the nitty gritty of study startup, site negotiation, enrollment, and data monitoring."
The bottom line: AI advocates have pushed back hard on the past week's safety debate, arguing that slowing the frontier would delay life-saving medicines. Rhetoric, however, is no substitute for results.
